Skip to main content

VMs as a service

Offer your tenants virtual machines on demand. A tenant picks a VM size from your catalog, and vMetal creates the VM on a KubeVirt cluster you operate, with the tenant's SSH keys on it.

This page shows how the pieces fit together and the order to build them in. Each step links to the guide that owns the procedure.

What the service provides​

  • A VM sized from your catalog. Each size is a node type that sets CPU, memory, disk, and image.
  • Self-service requests. Tenants request a VM from the Instances view in their project, or through the management API as a NodeClaim. The request form also lets tenants attach data volumes.
  • Their own guest. Each VM runs its own guest operating system and kernel. It boots your chosen image with the tenant's SSH keys and user data.
  • Clean release. Releasing the instance deletes the VM.

The same KubeVirt provider can also supply VMs that join clusters as private nodes, from the same catalog and the same API.

How it works​

vMetal exposes one stable Machine API upward and uses pluggable provisioning drivers underneath. For VMs, the driver is KubeVirt.

LayerWhat it doesWhere it's documented
KubeVirt clusterA connected control plane cluster that runs KubeVirt and hosts the VMs. The platform can deploy KubeVirt for you.Virtual machines overview
KubeVirt node providerConnects the platform to that cluster and holds the base VM template.KubeVirt node provider
Node typeOne VM size in your catalog. It can override or merge parts of the provider's template.Define node types
InstanceA tenant's claim on one VM, represented by a NodeClaim.Instance and server concepts
ProjectDecides who can request VMs, which sizes they can use, and how many they can hold.Allowed node types, Instance quotas

Build the service​

  1. Install vCluster Platform on the cluster that will run the management plane. Instances require the Machine Management feature in your license. See Install.

  2. Connect a cluster for your VMs and create a KubeVirt node provider. The provider can install KubeVirt on that cluster or use an existing installation. See Basic setup and the KubeVirt node provider reference.

  3. Define your VM catalog. Create a node type for each VM size. Then set up the images and networks the VMs use:

    Each node type's maxCapacity caps how many VMs of that size the provider creates.

  4. Give tenants access. Give each tenant one or more projects. Restrict each project to the VM sizes it may use, and cap how many instances it can hold. See Allowed node types and Instance quotas. Tenants reach their VMs through the platform, not through the KubeVirt cluster. See where the service ends.

  5. Let tenants request VMs. Tenants request, connect to, and release VMs as described in Independently provisioned instances. For the tenant-facing overview, see Instances. To test the service, request one VM from a tenant project and connect to it over SSH.

Operate the service​

  • Watch capacity. VMs consume CPU, memory, and storage on the KubeVirt cluster. Size each node type's maxCapacity to what that cluster can hold.

  • Manage VMs directly when you need to. The platform's virtual machine views let administrators start, stop, and restart VMs, open a console, and take snapshots. See Manage virtual machines.

    Experimental

    The virtual machine management views are experimental and enabled per customer through the VM Management feature. Don't rely on them for production operations without guidance from vCluster Labs. Tenant instances don't depend on them.

  • Troubleshoot storage and networking. See Troubleshoot data volumes and Troubleshoot network attachment definitions.

Know where the service ends​

vMetal owns the VM's creation, its first boot, and the claim on it. Your image and the tenant own everything inside the guest after that, including patching, monitoring, and any software the tenant installs.

VMs from different tenants can run on the same physical nodes of the KubeVirt cluster. Each VM has its own guest kernel, but tenants that need dedicated hardware need bare metal instead.

The KubeVirt cluster itself stays with you. A VM requested outside a cluster is created in the node provider's namespace on that cluster, so VMs from different tenants share that namespace. Tenants reach their VMs through the platform and over SSH. Don't give tenants direct access to the KubeVirt cluster.